Waverley builds the parts you never see.

Jibo had a face you could draw with one finger: a white circle on a black screen. The home robot could turn toward its owner and make a machine seem companionable. But consider everything that had to happen behind that circle. A phone needed to find the robot. A household needed to control its photographs. Servers needed to keep the conversation moving. Somewhere behind the personality was an engineering organization.

Waverley supplied part of that organization. Its Jibo work covered backend infrastructure, native iOS and Android applications, and quality assurance. This is a useful introduction to a company whose output often arrives wearing somebody else’s name. Waverley sells the expertise needed to connect the pieces of a product, including the pieces a buyer will never notice unless they stop working.

A robot is a whole system

For Jibo, Waverley describes a team of twelve specialists, divided across backend, mobile, and QA. The servers used independently scalable services packaged in Docker, with Amazon cloud infrastructure underneath. The mobile applications handled onboarding, family management, messaging, and robot-captured media. Encryption needed to work across devices while keeping access within the household.

The interesting point is the spread of responsibilities. Buying a robot means buying an ongoing relationship between hardware, software, and a remote service. A beautiful enclosure cannot settle a disagreement between an Android phone and a server. Waverley’s role was to work across those boundaries. That breadth is its most convincing distinction in a market crowded with companies offering to write code.

The Sidekick apprenticeship

The habit predates today’s AI vocabulary. Matt Brown founded Waverley in 1992; its account of his early work includes applications for Apple’s Newton. Later came Danger, the company behind the Hiptop, better remembered as the T-Mobile Sidekick. Danger needed engineering capacity without burdening an already stretched internal team.

Waverley sent a small Eastern European group to Danger’s Palo Alto offices. They learned the code alongside the customer’s engineers before the relationship expanded across product lines. Waverley calls the approach “crawl, walk, run.” The sequence matters. Bringing in outside developers creates a training bill payable in the existing team’s time. Starting small makes that bill easier to absorb.

“crawl, walk, run”

Waverley’s description of the Danger engagement

What earned the larger assignment, in Waverley’s telling, was visible competence and a quick ramp. There is a lesson here for the buyer tempted by a grand staffing proposal: give a small group a real piece of the system and watch how it learns. Expansion becomes a decision based on working evidence.

The bugs outside the checklist

Seagate’s problem was more specific. A network-attached storage launch needed testing capacity while internal QA resources were committed elsewhere. Waverley assembled three QA engineers and a lead in Vietnam. Its case study says the group reached productivity within a month, testing compatibility, storage behavior, and new releases.

Then comes the revealing detail: Waverley says most significant bugs discovered and fixed came from ad hoc testing. Experienced testers went looking beyond the written script. The immediate constraint was scarce QA capacity; the finding challenged the sufficiency of a checklist. For a buyer, the useful question becomes: which awkward combinations will this team know to try?

GridPoint offers a quieter example of learning traveling in the other direction. Waverley built mobile applications for its energy analytics platform, fitting tables and graphs onto a small screen. A “Choose Column” control let users prioritize the data they cared about. According to the case study, GridPoint subsequently adopted the mobile design language in its web product. The smaller screen had forced a useful decision.

Now the tutor needs plumbing

For Wall Street Prep, the obstacle was finding knowledge inside a growing training library. Waverley built Ask Ark, an AI assistant that retrieves platform content and generates responses linked to relevant passages, video timestamps, and audio markers. It began with core functionality and expanded into contextual follow-ups and recommendations.

The telling feature is evaluation: generated answers are scored against instructor-provided answers using an LLM judge. That creates a checking mechanism, though it should never be mistaken for a guarantee. The practical value is easy to understand. A finance student can get to the relevant explanation without manually hunting through a long lesson.

Waverley now presents Skywood as a foundation for agentic and conversational applications. Its pitch includes integration, guardrails, observability, and cost monitoring, with architectural patterns that can be implemented on different platforms. That places it among engineering partners helping businesses move AI into everyday operations. The familiar systems questions have acquired a new interface.

The price of knowing where to look

The commercial arrangement is services: extend a customer’s team or deliver a defined project, with ongoing support where needed. Its public Clutch listing shows a $50,000-plus project minimum and a $50-$99 hourly range. Those figures are purchasing signposts, not the budget for Jibo or Ask Ark. A connected product’s cost depends on the scope of all those interlocking responsibilities.

There are limits to the fit. A simple requirement already served by standard software may not justify custom development. A customer unable to explain its system or make product decisions will make onboarding harder. And a buyer choosing entirely on hourly rate is unlikely to value Waverley’s stated emphasis on senior judgment.

The methods are more portable than the contract: start with a bounded assignment, involve testers early, and measure AI answers against relevant expertise. Waverley’s careers materials emphasize ownership and learning; its customer accounts show where those habits become useful. Behind the circle on the robot’s face, somebody still has to understand the whole machine.